columnstore is supposed to not work with indices (although they have different types like zone-maps)
each column is stored separately on disk, so only the requested column-values are read from disk
This makes slow to select/update/delete a single row(oltp) since it needs to fetch multiple pages (for each column).
And makes it fast to do queries regarding data in big size (olap) by doing vectorized query execution and sequential-reads on disk
Note, updatetable columnstore sucks everywhere, so I don't think so. It's only for OLAP, not OLTP (parse is oltp).
What may help @azinman is group-compression (ex: bigger pages of rows and compress the whole page, where each json-key can be repeated multiple times in a page since each page may have multiple rows). This is what happens on tokudb,hbse,hypertable,cassandra.